Twice in the last year. Our infrastructure is all on UPS, but we decided not to do so for our compute cluster. Additional hardware was deemed a better way to

Twice in the last year. Our infrastructure is all on UPS, but we decided not to do so for our compute cluster. Additional hardware was deemed a better way to spend our budget. In the event of an outage, people lose the job that was currently running. All their data is safe. They just need to relaunch the job and they are back in business. Most jobs take just a few days, so it really isn't that big of a deal. Either users wait in a queue for an extra day every time they want to run a job, or once or twice a year they have to resubmit a job that died in a power outage. We chose the latter.